The aspects described herein relate to photonic optical flow switches.
An optical flow switch or an optical circuit switch (OCS) is a basic building block in high-performance data centers or computing systems and routes optical signals between portions of the data center or computing system. Optical flow switches may additionally route optical signals that are encoded in either the amplitude, phase, directionality, and/or polarization of light. Optical flow switches can be used to adapt the network topology such that, for example, servers that communicate high bandwidth data can be awarded additional direct optical connections to increase their bandwidth.
Some embodiments provide for an optical flow switch. The optical flow switch comprises a network of photonic switches arranged between input and output ports of the optical flow switch, and the network of photonic switches spans two or more reticles.
In some embodiments, the two or more reticles comprise a same reticle pattern, and the network of photonic switches is formed of a repeating arrangement of the same reticle pattern.
In some embodiments, the network of photonic switches comprises a Cantor network or a Benes network.
In some embodiments, the network of photonic switches comprises a first plurality of switches arranged to form Benes switches.
In some embodiments, the network of photonic switches further comprises a second plurality of switches comprising input routing switches coupled between the input ports and the Benes switches. In some embodiments, switches of the second plurality of switches are coupled between an input port of the input ports and each Benes switch of the Benes switches. In some embodiments, switches of the second plurality of switches are coupled between an input port of the input ports and Benes switches disposed across the two or more reticles.
In some embodiments, the second plurality of switches comprise: first input routing switches arranged to receive optical signals from the input ports, second input routing switches arranged to receive optical signals from the first routing switches, and third input routing switches arranged to receive optical signals from the second routing switches and to transmit optical signals to the Benes switches.
In some embodiments, the Cantor network comprises an N×N Cantor network, the input ports comprise k input ports, the two or more reticles comprise r reticles, and the Benes switches comprise m total Ñ×Ñ Benes switches, where Ñ=2ceil(log
In some embodiments, the first input routing switches comprise k total 1:r photonic switches; the second input routing switches comprise kr total 1:r photonic switches; and the third input routing switches comprise kr2 total r:q photonic switches.
In some embodiments, the network of photonic switches further comprises a third plurality of switches comprising output routing switches coupled between the Benes switches and the output ports. In some embodiments, switches of the third plurality of switches are coupled between each Benes switch and an output port of the output ports. In some embodiments, switches of the third plurality of switches are coupled between Benes switches disposed across the two or more reticles and an output port of the output ports.
In some embodiments, the two or more reticles comprise rows or columns of reticles, each row or column of reticles being formed using a same set of reticle patterns.
In some embodiments, the rows or columns of reticles comprise X rows or columns of reticles, and the optical flow switch further comprises 1:X and/or X:1 switch routers disposed between photonic switches of the network of photonic switches. In some embodiments, X=2. In some embodiments, X=4.
In some embodiments, the optical flow switch further comprises 1:2 and/or 2:1 switch routers disposed between photonic switches of the network of photonic switches.
In some embodiments, photonic switches of the network of photonic switches that are disposed in a first reticle are coupled to photonic switches of the network of photonic switches disposed in a second reticle by waveguides.
In some embodiments, photonic switches of the network of photonic switches are disposed in an arrangement spanning two or more chips. In some embodiments, the two or more chips are coupled using optical fibers.
In some embodiments, photonic switches of the network of photonic switches comprise a 2×2 Mach-Zehnder interferometer (MZI) switch. In some embodiments, photonic switches of the network of photonic switches comprise three MZI switches arranged in series. In some embodiments, photonic switches of the network of switches comprise a dilated MZI switch. In some embodiments, photonic switches of the network of photonic switches comprise silicon photonic switches.
In some embodiments, the input and output ports comprise grating couplers, edge couplers, and/or v-groove couplers.
Some embodiments are directed to an optical switch package. The optical switch package comprises a switch photonic integrated circuit (PIC) comprising an optical flow switch and a glass substrate supporting the switch PIC.
In some embodiments, the optical switch package further comprises a plurality of input ports and a plurality of output ports, each of the plurality of input ports and each of the plurality of output ports comprising active pluggable fiber array units. The active pluggable fiber array units comprise a plurality of optical couplers configured to optically couple optical fibers to the glass substrate, and a plurality of optical amplifiers configured to amplify optical signals at each of the plurality of optical couplers.
In some embodiments, the optical switch package further comprises glass waveguides optically coupling the active pluggable fiber array unit to the switch PIC. In some embodiments, the glass waveguides are a fanout between the switch PIC and the active pluggable fiber array. In some embodiments, the glass waveguides are coupled to the switch PIC by a plurality of evanescent couplers.
In some embodiments, the switch PIC further comprises control circuitry coupled to the optical flow switch.
In some embodiments, the control circuitry is monolithically integrated in the switch PIC.
In some embodiments, the optical switch package further comprises a first amplifier PIC and a second amplifier PIC. The first amplifier PIC optically couples the input ports to the switch PIC, and the second amplifier PIC optically couples switch PIC and the output ports.
In some embodiments, the switch PIC comprises a first switch PIC and a second switch PIC, the first switch PIC and the second switch PIC comprise substantially identical switch networks, and the second switch PIC comprises an optical flow switch having a switch network that is rotated 180 degrees relative to a switch network of the first switch PIC.
In some embodiments, the optical switch package further comprises a third amplifier PIC optically coupled between the first and second switch PICs. In some embodiments, the optical switch package further comprises a first amplifier PIC, and a second amplifier PIC, wherein: the first amplifier PIC optically couples the input ports and the first switch PIC, and the second amplifier PIC optically couples the second switch PIC and the output ports.
In some embodiments, the glass substrate comprises a first glass substrate and a second glass substrate, the first switch PIC is disposed on the first glass substrate, and the second switch PIC is disposed on the second glass substrate.
Some embodiments are directed to a method of manufacturing an optical switch package. The method comprises transferring two or more reticles comprising at least one reticle pattern to a photoresist supported by a switch photonic integrated circuit (PIC) substrate, depositing an optical material on the switch PIC substrate through the portions of the photoresist corresponding to the transferred two or more reticles, wherein depositing the optical material forms at least part of an optical flow switch comprising a network of photonic switches spanning the two or more reticles, coupling a chip comprising control circuitry to the switch PIC substrate, and bonding the switch PIC substrate to a glass substrate.
In some embodiments, transferring the two or more reticles comprises transferring two or more reticles comprising a same reticle pattern to the photoresist such that the network of photonic switches is formed of a repeating arrangement of the same reticle pattern.
In some embodiments, depositing the optical material forms a network of photonic switches comprising a Cantor network or a Benes network.
In some embodiments, depositing the optical material forms a network of photonic switches comprising a first plurality of switches arranged to form Benes switches.
In some embodiments, depositing the optical material further forms a second plurality of switches comprising input routing switches coupled between input ports disposed on the switch PIC substrate and the Benes switches.
In some embodiments, depositing the optical material further forms waveguides coupling switches of the second plurality of switches to Benes switches disposed across the two or more reticles.
In some embodiments, transferring the two or more reticles comprises transferring rows or columns of reticles, each row or column of reticles being formed using a same set of reticle patterns.
In some embodiments, transferring the two or more reticles comprising at least one reticle pattern to a photoresist comprises transferring the two or more reticles to photoresist supported by two switch PIC substrates.
In some embodiments, depositing the optical material comprises depositing silicon.
In some embodiments, the method further comprises bonding a first amplifier PIC and a second amplifier PIC to the glass substrate, wherein: the first amplifier PIC optically couples input ports of the glass substrate to the switch PIC substrate, and the second amplifier PIC optically couples the switch PIC substrate to output ports of the glass substrate.
In some embodiments, bonding the switch PIC substrate to the glass substrate comprises bonding two switch PIC substrates to the glass substrate, the two switch PIC substrates supporting substantially identical networks of photonic switches.
In some embodiments, bonding the two switch PIC substrates to the glass substrate comprises bonding a first one of the switch PIC substrate to the glass substrate and bonding a second one of the switch PIC substrate to the glass substrate with a rotation of 180 degrees in a plane parallel to a plane of the glass substrate relative to an orientation of the first switch PIC substrate.
In some embodiments, bonding an amplifier PIC to the glass substrate such that the amplifier PIC is optically coupled between the first and second switch PIC substrates.
In some embodiments, the glass substrate comprises a first glass substrate and a second glass substrate, bonding the first switch PIC substrate to the glass substrate comprises bonding the first switch PIC substrate to the first glass substrate, and bonding the second switch PIC substrate to the glass substrate comprises bonding the second switch PIC substrate to the second glass substrate.
Various aspects and embodiments will be described with reference to the following figures. It should be appreciated that the figures are not drawn to scale. In the drawings, each identical or nearly identical component that is illustrated in various figures is represented by a like numeral. For purposes of clarity, not every component may be labeled in every drawing.
Optical flow switches are typically implemented using three-dimensional free-space micro-electromechanical systems (MEMS) mirror switches. These MEMS mirror switches can scale up to approximately 1000 input and output ports with switching times on the order of 10-20 ms. However, the fabrication of optical flow switches using three-dimensional MEMS mirror switches is challenging and requires calibration and installation of discrete components such as two-dimensional fiber arrays, micro-lens arrays, and beam steering mirrors. Furthermore, optical flow switches typically rely on the concurrent use of electronic packet switches which results in packet drops during periods of traffic contention. The high production costs, relatively slow switching speeds, and potential data loss discourage the implementation of three-dimensional MEMS optical flow switches in large computational systems and data centers.
The inventors have recognized and appreciated that the use of integrated photonics (e.g., silicon photonics) in optical flow switches offers a path towards implementing optical flow switches in computing systems and data centers by providing miniaturization and improved phase stability of optical signals transmitted through the optical flow switches. Additionally, the use of integrated photonics provides significantly faster (e.g., approximately 1 μs or less) switching times. This faster switching time lends itself to better programmability of the optical flow switches and better utilization. For example, the switches can be reconfigured during runtime, and endpoints of the switches can be linked in a round-robin fashion during runtime to improve utilization.
The inventors have further recognized and appreciated that integrated photonics are implemented on chips using relatively few layers for routing the optical signals (e.g., silicon and silicon nitride layers in the case of silicon photonics). Thus, the routing of optical signals in integrated photonics systems is inherently two-dimensional. Such two-dimensional photonic switch structures have conventionally limited the number of input and output ports because of reticle-size limits of the fabrication masks for photolithography. Additionally, photonics switching structures have been limited in size due to the difficulty in controlling and stabilizing photonic switches simultaneously.
Accordingly, the inventors have developed multi-reticle photonic optical flow switches and photonic switching structures that have low manufacturing costs (e.g., being manufacturable with 200 mm and 300 mm complementary metal-oxide-semiconductor (CMOS) processes), are scalable to large numbers of input and output ports, and have fast reconfiguration times on the order of sub-microseconds. Additionally, the multi-reticle photonic optical flow switches described herein enable non-blocking optical flow switches with mode sizes larger than those achievable with single-reticle optical flow switches.
In some embodiments, the multi-reticle optical flow switches described herein includes a series of substantially identical adjacent reticles that together implement a network of switches, as shown in the example of
In some embodiments, the network of switches forming the multi-reticle optical flow switch 100 includes photonic switches arranged in the Cantor network topology. In a general N×N Cantor network, a set of N input optical fibers is coupled to the multi-reticle optical flow switch 100 by input ports 104a and routed-via input routing switches 106a—to an intermediate switch layer including log(N) independent Ñ×Ñ Benes switches, where N=2ceil(log
Spatial dilation across the log(N) Benes switches 108 introduces a circuit layout complexity that is prohibitive for large N for fabrication using a single two-dimensional reticle. In some embodiments, the multi-reticle optical flow switch 100 distributes the log(N) Benes switches across multiple reticles (e.g., reticles 102a-102d), lowering the circuit complexity in each reticle. Each Benes switch 108 is independent and operates using waveguide connections within its reticle.
In some embodiments, the Benes switches 108 are implemented using a recursive structure, as illustrated in the example of
In some embodiments, the input and output routing switches 106a, 106b include routing of optical signals across reticle boundaries. The individual 2×2 photonic switches in different reticles may be connected by waveguides that extend across reticle boundaries, and waveguide continuity at the reticle boundaries may be implemented using reticle stitch technology.
In some embodiments, the input routing switches 106a, the Benes switches 108, and/or the output routing switches 106b are formed of photonic switches, an example of which is depicted in
In some embodiments, the input and output ports 104a, 104b may be coupled to multiple optical fibers. The optical fibers may be attached to the switch network using coupling techniques such as, for example, grating couplers, edge couplers, or v-groove couplers. A waveguide-array to fiber transposer (WAFT) may be used to reduce the pitch between the neighboring inputs from 127 μm or 250 μm for standard optical fibers to a pitch that is less than 10 μm. A WAFT can therefore be used to increase the number of spatial mode inputs and outputs at the reticle edge of the switch, which is currently limited to 32 mm by the size of standard 26 mm×32 mm reticles. Multi-core fibers may be attached to the input and output ports 104a, 104b using grating couplers that match the array of cores, thereby further increasing the number of spatial modes in and out of the switch.
It should be appreciated that while the examples of multi-reticle optical flow switches described herein are depicted as providing unidirectional communication, the described photonic switch networks can be used for bidirectional communication, as aspects of the technology described herein are not limited in that respect. For example, the endpoints on the left sides of the illustrative multi-reticle optical flow switches can be configured to send data to the endpoints on the right, and endpoints on the right can also be configured to send data to the endpoints on the left.
In some embodiments, the total number of Benes switches in the multi-reticle optical flow switch is given by m=qr, where each reticle is populated with q independent Ñ×Ñ Benes switches. The per-reticle switch count q may be chosen as the minimum integer such that qr≥log2Ñ. The number of optical fibers, k, at the input and output of each reticle is then given by the largest integer satisfying kr≥Ñ.
In some embodiments, input routing and/or output routing from the Benes switches of the multi-reticle optical flow switch may be implemented using three sets of photonic switches, as depicted in the example of a reticle 300 depicted in
In some embodiments, the first input routing switches 302 may be configured to provide optical outputs to second input routing switches 304. The reticle 300 may include kr total second input routing switches 304, which may be 1:r Benes Switch Select A (SSA) active switches. The second input routing switches 304 may control the reticle to which input data will be sent during transmission through the multi-reticle optical flow switch.
In some embodiments, the second input routing switches 304 may be configured to provide optical outputs to third input routing switches 306. The reticle may include kr2 total third input routing switches 306, which may be r:q Switch Select B (SSB) active switches. The third input routing switches 306 may be configured to control the routing of the optical signal to particular Benes switches 308-1 through 308-q within the reticle 300.
In some embodiments, output routing may be implemented with three sets of output routing switches that are structurally symmetric to the input routing switches 302, 304, and 306. For example, the output routing switches may include first output routing switches 310, which may include kr2 total q:r SSB active switches. The output routing switches may further include second output routing switches 312, which may include kr total r:1 SSA active switches. The output routing may also include third output routing switches 314, which may include k total r:1 RS active switches. The third output routing switches 314 may be optically coupled to k total outputs 315-1 through 315-k, each of which may be coupled to output optical fibers.
In some embodiments, the only connections between routing switches that are formed of reticle-crossing waveguides are the connections between the SSA and SSB switches. As each reticle in the multi-reticle optical flow switch is fabricated using identical fabrication masks, it is important to configure the inter-reticle waveguides to be continuous at the reticle boundary. The waveguide routing between SSA and SSB switches for a switch network distributed over r reticles {R1, R2, . . . , Rr} can be configured by instead considering a network with 2r−1 reticles {R−(r−2), R−(r−3), . . . , R0, R1, . . . , Rr}, where the reticle mask is located in the middle position at R1. An example of the waveguide connections between SSA switches 304 and SSB switches 306, using the extended reticle network for routing, are depicted in the example of
The multi-reticle input and output routing strategies described herein are also applicable to other switch networks in addition to the Cantor network. In the most general case, the reticle select (RS) stage (e.g., first input routing switches 302) may configure the switch network to perform the function of a specific reticle location. For example, the RS switches can be configured to select which input mode in the schematic a specific input optical fiber represents. The switch select A (SSA) stage (e.g., second input routing switches 304) may route the optical signal to the reticle that the destination switch is on (e.g., acting as an inter-reticle transmitter). The switch select B (SSB) stage (e.g., third input routing switches 306) may receive the signal routed by the SSA switches (e.g., acting as an inter-reticle receiver) and selects the intra-reticle destination Benes switch that the signal should be routed to. For a 2k Cantor network, the destination switch is a 2k Benes switch, and for a 2k Benes network, the destination switch is a 2k−1 Benes switch.
More generally, the input and output layers of a 2k×2k Benes network are composed of 2k-1 total 2×2 photonic switches. For the example of two reticles, the first input routing switch layer may be implemented with 2k-1 1:2 switches that specify which input a given fiber mode represents. The second input routing switch may include of 2k−1×2×2 switches, each receiving data from two switches of the first input switch layer simultaneously. The third input switch layer may include 2k−1 2:1 switches that each receive input from the second input routing switch layer and output to the local (e.g., on the same reticle)2k−1×2k−1 Benes switch.
Additionally, the routing strategies described herein may also be applicable recursively. For example, the strategy allows for the construction of a 2k Cantor network with destination switches that are 2k Benes switches. Each of the 2k Benes switches can be constructed using the same input and output routing strategies described in connection with
In some embodiments, data routing between reticles may be achieved using a variety of physical mechanisms. However, reticle stitch waveguide couplers expand the optical mode near the reticle boundary. Thus, low-loss transmission of data across a reticle boundary can be achieved in a continuous waveguide with two reticle stitch couplers aligned on either end of the reticle boundary. Reticle stitching enables monolithic switch fabrics composed of multiple identical reticles, without requiring dicing and waveguide coupling across independent chips.
The inventors have further recognized that the compressive strain experienced by commercial photonic wafers imposes limitations on the aspect ratio of monolithic photonic integrated circuits. In cases where a high aspect ratio is desired—for example, where the number of reticles r becomes large—the full switch network may be subdivided across a plurality of chips each having smaller monolithic sub-circuits composed of fewer reticles, in some embodiments. These sub-circuits may then be connected between chips using off-chip waveguide coupling approaches, such as fiber edge coupling, grating coupling, or photonic wirebonding.
In some embodiments, a hybrid coupling architecture may still be implemented using a single fabrication mask. Flexibility can be achieved in the implemented coupling mechanisms by either designing the reticle stitch coupler to exhibit mode-matching to an edge-coupled fiber or using an active switch to select the coupling mechanism.
III. Photonic Switches with Minimized Crosstalk
Crosstalk between optical singles in an optical flow switch can degrade the signal-to-noise ratio (SNR) of the transmitted optical signals. For example, if the optical flow switch is implemented using 2×2 switches having ∈s crosstalk (in percent) and optical crossings with ∈x crosstalk, then the SNR of the optical flow switch will be degraded by approximately (1−∈s)NS(1−∈X)NX, where NS and NX are the number of 2×2 switches and the number of optical crossings in one specific optical path respectively.
The inventors have recognized that the crosstalk of an optical flow switch is dominated by the crosstalk of the switches, which have a finite extinction ratio. In a 2×2 MZI switch, as shown in the example of
In some embodiments, reducing the crosstalk of the switch can be done by using multi-mode interferometers (MMI) as the input and output 50:50 splitters. MMIs are generally more resistant to deviations due to fabrication imperfection. Additionally, these MMIs can be designed using photonic inverse design which would generally perform better with finer lithographic definition.
In some embodiments, and as shown in the example of
In some embodiments, and as shown in the example of
Reducing crosstalk may also be achieved by reducing crosstalk in the optical crossings. For example, improved crossing structures with reduced crosstalk may be created using photonic inverse design and fabricated with better lithographic precision allows for finer patterning. Another way of reducing this crosstalk in crossing structures is to use more than a single optical routing layer. A crossing between two silicon and silicon-nitride layers spaced at approximately 2 μm apart vertically will lead to less than 1 mdB loss and less than −80 dB crosstalk per crossing. At this vertical distance, an intermediary layer (e.g., a silicon nitride layer) may be used to send the signal from the bottom silicon layer to the top silicon-nitride layer. In such an arrangement, an optical signal is first transferred to the intermediate layer through a taper and thereafter the signal is transferred to the top layer through another taper. Transferring the signal from the top layer to the top layer is also achievable through the intermediate layer. Lower crossing loss and crosstalk are achievable by increasing the distance between the silicon and the silicon-nitride layers. At larger vertical distances, more than a single intermediate layer may be used.
Multiple modulation physics schemes are suitable for the photonic switching elements. For example, carrier injection or depletion in silicon photonic devices can be used to modulate the switches between the cross and the bar states. While carrier injection and depletion allows for sub-picosecond modulation, the mechanism exhibits phase-dependent loss. As another example, physically actuating the arms of the 2×2 MZI switches using micro-electromechanical systems (MEMS) can also modulate the switches between the cross and the bar states. MEMS actuation can provide modulation in the sub-microsecond regimes. The process of releasing the MEMS switches, however, may not always be compatible with the fabrication process used to manufacture integrated photonics (e.g., CMOS-compatible silicon photonic processes). As a further example, in a system implemented using non-linear photonic materials (e.g., χ(2) in InP), the switches can be modulated using non-linear optics. The non-linear process can be used to provide low-loss phase shifts to modulate the switch between the cross and bar states. If the non-linear process is sufficiently strong (e.g., through large amplifications), one optical signal can modulate a switch acting on another optical signal. Finally, thermo-optic modulation is another modulation technique that is compatible with the photonic switching elements. Heating up or cooling down the waveguides of the switching element effectively induces a phase shift which can be used to modulate the switch between the cross and the bar states.
In all the examples described above, equalization techniques can be used to shorten the switching time between the bar and the cross states, and vice versa. An optimal excitation signal can be derived by compensating for the slower responses limited by the physics of the modulation. For example, in a thermo-optic modulation, a pre-emphasis filter that takes into account the relatively slow speed of heat propagation and diffusion processes can induce a switching response that is faster than a simple step response time. The pre-emphasis filter may be designed to increase the magnitude of some frequency responses (e.g., high frequency responses). Similarly, a different pre-emphasis filter can take into account carrier recombination time in switching performed using carrier-depletion, or a different pre-emphasis filter can take into account the slower actuation bandwidth in MEMS devices in MEMS switching techniques.
In some embodiments, the equalization filters can be applied in a single-ended scheme or a dual-ended scheme. Other equalization schemes include a de-emphasis filter which attenuates the magnitude of some frequency responses. De-emphasis filters, for example, can be used to counteract the high-frequency losses of some modulation schemes (e.g., carrier injection or depletion) to achieve a flat modulation response.
In some embodiments, the analog front end driving the switches and maintaining the stability of the switches (e.g., against temperature fluctuations during the operations) can be integrated monolithically in the same chip (e.g., for silicon photonic switches), integrated using a 3D stacking approach (e.g., application-specific integrated circuits (ASICs) with analog output DACs and readout ADCs bonded on top of the switch with direct bump-to-bump connections or through an interposer), or integrated using a two-dimensional or two-and-a-half-dimensional approach (e.g., the ASIC sends and receives its signals through an interposer substrate or through a PCB).
The inventors have further recognized that it may not be efficient or possible to reconfigure the optical flow switch network at nanosecond or picosecond time scale. The switch network may instead be configured on an application-by-application basis. The optical flow switch may be configured to create a network topology between the end-point processor and memory units that enables the best runtime performance of the computing system for a particular application before runtime. When another application is to be run in the system, the optical flow switch may be reconfigured to another network topology that enables the best runtime performance for that second application. In this scenario, the optical flow switch may be quasi-static.
In some embodiments, the optical flow switch may be dynamic even at slower times than the clock frequencies (e.g., on the order of GHz). In one example, one or more end-points of the switch system can request the control circuitry controlling the optical flow switch to provide one or more optical connections from one transmitter end-point to one or more receiver end-points. The number of optical connections may depend on the requested bandwidth. For example, if the optical communication transceiver uses eight wavelengths of light per connection, with each wavelength modulated at 56 Gbps, then an optical connection provides 448 Gbps between a transmitter and a receiver. Two optical connections then provide double the bandwidth (e.g., 896 Gbps) between a transmitter and a receiver. The request for additional optical connections may be made on top of a packet switch network, where if a request is received to send a large amount of data, then optical connections are provided within the optical switch network. Otherwise, data may simply be transmitted through the packet switch network electrically.
In some embodiments, the connections between the sending and the receiving end-points can be set up with a pre-determined schedule. For example, the senders and the receivers can be paired up in a round-robin manner such that each sender has the opportunity to send data to each receiver during their pre-determined schedule. The pre-determined scheduled is stored in the memory of the senders, the receivers, and in the optical switch network such that every participating element knows, at any period of time, who the senders are and who the receivers are. In some embodiments, the pre-determined schedule may be stored within the optical switch network. The sender and the receiver can identify themselves before any data is transmitted.
In some embodiments, the connection can also be made dependent on the wavelength transmitted from the sender. In a wavelength multiplexed scheme, the input data can be demultiplexed using a collection of resonant devices (e.g., ring resonators or racetrack resonators) or a series of unbalanced MZIs. The path taken by each wavelength can be different or the same. Data can also be wavelength multiplexed back into a single output waveguide or fiber using the same resonant devices. In another scenario, one or more wavelengths can be used to communicate the switch network configuration to the control processor.
In some embodiments, the optical flow switch may be configured for broadcasting. The 2×2 photonic switches may be configured to split an input signal to any ratio between the bar state (100:0) and the cross state (0:100). Using this property, the optical flow switch can be configured to broadcast a single input signal, or multiple input signals in different wavelengths multiplexed into a single input waveguide, to more than one receiver.
One or more of the optical flow switches 1010 may include an amplifier 1012 coupled adjacent to inter-switch optical couplers 1014, the amplifier 1012 may be configured to overcome losses introduced by the optical flow switches 1010. The amplifiers 1012 can be fabricated within the same monolithic circuit as the optical flow switches 1010 or be fabricated separated from the optical flow switches 1010 and positioned before or after one or more of the optical flow switches. More than a single amplifier can also be placed in the system. Where the amplifiers should be placed depend on the minimum signal power required by the amplifiers, the maximum saturated output of the amplifiers, as well as the desired signal power received at the receiver. The amplifiers 1012 may be semiconductor optical amplifiers (SOAs), erbium-doped fiber amplifiers (EDFAs), bismuth-doped fiber amplifiers (BDFAs), or praseodymium-doped fiber amplifiers (PDFAs).
In some embodiments, the photonic interconnect fabric 1106 include built-in circuit switching that allows for reconfiguration of communication topology between ASICs, memory chips, and/or network chips that are using the photonic interconnect fabric 1106. For example, one or more of these chips may be coupled on top of each site of the photonic interconnect fabric 1106 and the electronic communications signals output by one or more of these chips may be converted to photonic signals within the photonic interconnect fabric 1106.
In some embodiments, the photonic interconnect fabric 1106 can be configured to provide all-to-all connections between the electronic chips disposed on top of the photonic interconnect fabric 1106. Additionally, each site of the photonic interconnect fabric 1106 has one fiber input and output to the optical flow switch 1104 within the same rack 1102a-1102d. Some of the outputs of one optical flow switch 1104 can be connected to the inputs of the other optical flow switches 1104.
VIII. Multi-Reticle Optical Flow Switches with Multiple Reticle Mask Patterns
The inventors have further recognized and appreciated that certain photonic switch architectures may benefit from being manufactured using more than a single reticle mask pattern. As an example, consider the problem of constructing a 16×16 Benes switch 1300 comprising 56 photonic switches 1302, as shown in the example of
In contrast, the example of
In the example of
In contrast, and as shown in the example of
This routing strategy described above may be scaled up to accommodate larger N×N Benes switches.
In some instances, it may be preferable to, or required by the fabrication facility, to divide the switch network between multiple columns of reticles, where each column use the same reticle mask set. In a Benes or Cantor network switch, there is a symmetry around the center of the switch. The symmetry can be described as either a mirror symmetry or a 180-degree rotation around the z-axis. If the fabrication process allows for either of these symmetries, then the optical flow switches as described in
In some embodiments, for an optical signal to traverse the row of reticle 1811a and 1811b from left to right, it may enter the switch network at edge couplers 1812a. The optical signal may then encounter a first pair of switches 1814. The first pair of switches 1814 may be configured to select whether the input optical signal will pass through the first pair of switches 1814 to the remaining switches within the reticle 1811a or if the input optical signal will be output from the reticle 1811b by edge couplers 1812b.
If the first pair of switches 1816 are configured to pass the optical signal from the first pair of switches 1814 to the remainder of the optical flow switch, then the optical signal will be transmitted to the second pair of switches 1816. In some embodiments, the second pair of switches 1816 may be configured to select whether the received optical signal will be output to the next reticle 1811b, or if the reticle 1811b will instead receive inputs from a reticle (not shown) to the left of the reticle 1811a.
The inventors have also appreciated that the switch routing techniques described in connection with
It should be appreciated that while the switch network design techniques described in connection with
In some embodiments, an optical flow switch (e.g., formed using one reticle or formed using multiple reticles, as described herein) can be packaged using a glass substrate. In the examples of
In some embodiments, the fan out allows for the use of direct pluggable fiber array connectors 2004 comprising, for example, multi-fiber connectors (e.g., a 16-fiber or a 32-fiber multifiber push-on (MPO) connector) that can be directly plugged into the glass substrate to optically couple optical fibers 2003a and 2003b to the glass waveguides 2009. The direct pluggable fiber array connectors 2004 may be active pluggable fiber array units. For example, the direct pluggable fiber array connectors 2004 may include semiconductor optical amplifiers (SOAs) for amplifying the optical signals entering and exiting the optical flow switch package 2000 so that overall loss caused by the optical flow switch package 2000 may be minimized.
In some embodiments, the glass substrate 2002 also provides the electrical connections between the switch PIC 2010 and the substrate 2001. The glass substrate may include a bottom redistribution layer (RDL) 2006 and a top RDL 2008. The glass substrate 2002 may be bonded to the substrate 2001 by electrical connections 2005 (e.g., by solder-bonding or other suitable electrical bonding techniques) formed between the substrate 2001 and the bottom RDL 2006. The bottom RDL 2006 may be electronically coupled to the top RDL 2008 by through-glass vias 2007.
In some embodiments, the switch PIC 2010 is bonded to the glass substrate 2002 at top RDL 2008 (e.g., by solder-bonding or other suitable electrical bonding techniques). The switch PIC 2010 includes photonic switch elements 2016 (e.g., photonic switch networks forming optical flow switches, as described herein). Additionally, a control chip 2020 including control circuitry 2022 is bonded on top of the switch PIC 2010. The control circuitry 2022 may be configured to provide control signals to the photonic switch elements 2016 (e.g., to reconfigure the photonic switch elements 2016 during operation of the optical flow switch package 2000). Electrical signals from the substrate 2001 are routed to the control chip 2020 using through silicon vias (TSVs) 2018.
In some embodiments, amplification may be provided by one or more PICs disposed on the glass substrate 2002 rather than by active fiber array units.
In some embodiments, the mirror symmetry of the Benes or Cantor switch network may be utilized to simplify fabrication of the switch PIC.
In some embodiments, additional amplification between the two switch PICs 2210a and 2210b may be provided by direct pluggable fiber array connectors 2004.
In some embodiments, the amplification at the input and output of the optical flow switch may be provided by direct pluggable fiber array connectors 2004.
In some embodiments, the two or more reticles have a same reticle pattern. When transferring the two or more reticles to the photoresist, the same reticle pattern may be stamped repeatedly across the photoresist, iteratively exposing the photoresist to transfer a repeating reticle pattern that may be used to form the network of photonic switches. Alternatively, the two or more reticles may be associated with sets of reticle mask patterns. The sets of mask patterns may be stamped repeatedly across the photoresist, iteratively exposing columns or rows of the photoresist to transfer repeating reticle patterns in columns or rows that may be used to form the network of photonic switches.
In some embodiments, the network of photonic switches may be formed across two switch PIC substrates. Transferring the two or more reticles to the photoresist may then comprises transferring the two or more reticles to regions of photoresist that correspond to regions that will be used to form separated switch PIC substrates. For example, the two or more reticles may be transferred to regions of the photoresist on portions of the substrate that will be diced apart from one another. Alternatively, the two or more reticles may be transferred to photoresists located on separate substrates.
After act 2402, process 2400 may proceed to act 2404 in some embodiments. At act 2404, an optical material may be deposited on the switch PIC substrate through the portions of the photoresist corresponding to the transferred two or more reticles. The optical material may be deposited by any suitable deposition technique (e.g., chemical vapor deposition, atomic layer deposition, etc.). The optical material may comprise, for example, a semiconductor (e.g., silicon).
In some embodiments, depositing the optical material forms at least part of a network of photonic switches comprising a Cantor network or a Benes network. In some embodiments, depositing the optical material forms at least part of a network of photonic switches comprising a first plurality of switches arranged to form Benes switches. Depositing the optical material may also form a second plurality of switches comprising input routing switches coupled between input ports disposed on the switch PIC substrate and the Benes switches. Finally, depositing the optical material may form waveguides and/or waveguide crossing coupling switches of the second plurality of switches to Benes switches disposed across the two or more reticles.
After act 2404, process 2400 may proceed to act 2406, in some embodiments. At act 2406, a chip comprising control circuitry may be coupled to the switch PIC substrate. For example, the chip may be electrically coupled to a top surface of the switch PIC substrate (e.g., as shown in the example of
After act 2406, process 2400 may proceed to act 2408, in some embodiments. At act 2408, the switch PIC substrate may be bonded to a glass substrate. For example, the switch PIC substrate may be electrically and optically coupled to a top surface of the glass substrate (e.g., as shown in the example of
In some embodiments, bonding the switch PIC substrate to the glass substrate also includes bonding a second switch PIC substrate. The two switch PIC substrates may support substantially identical networks of photonic switches, each switch PIC substrate supporting half of a complete switch network used to form the optical flow switch (e.g., as described in connection with
In some embodiments, the process 2400 further includes bonding one or more amplifier PICs to the glass substrate. For example, two amplifier PICs may be bonding to the glass substrate, a first amplifier PIC coupling input ports of the glass substrate to the switch PIC substrate and a second amplifier PIC coupling the switch PIC to output ports of the glass substrate. In some embodiments including two switch PICs, alternatively or additionally, an amplifier PIC may be bonded to the glass substrate between the two switch PICs such that the two switch PICs are optically coupled by the amplifier PIC.
In some embodiments, the glass substrate includes two glass substrates and the switch PIC includes two switch PICs. Bonding the switch PIC to the glass substrate may then include bonding a first switch PIC to a first glass substrate and bonding the second switch PIC to a second glass substrate. In some embodiments, the method may further include optically coupling the two switch PIC substrates using, for example, optical fibers coupled between the two glass substrates.
In some embodiments, the optical flow switch may be configured to enable bidirectional communication using two different polarizations of light within single optical fibers.
In some embodiments, the amplifier and polarization demultiplexing PICs 2340a and 2340b are each configured to demultiplex incoming optical signals and multiplex outgoing optical signals using the polarization state of the light. If the optical fibers coupled to the passive fiber array units 2104 are polarization-maintaining (PM) optical fibers, then the amplifier and polarization demultiplexing PICs 2340a and 2340b include a polarization splitter-rotator module that receives as input two signals in two polarizations (e.g., TE and TM polarizations) within a single glass/silicon waveguide and outputs to two separate waveguides using TE modes only. If the optical fibers coupled to passive fiber array units 2104 are single-mode (SM) optical fibers, then the amplifier and polarization demultiplexing PICs 2340a and 2340b include polarization devices as described in connection with
The arrangement of optical flow switch package 2300 allows for communication links to be arranged at the simplex level (e.g., from one transmitter to only one receiver). For example, the transmitter of a first PIC may be in communication with the receiver of a second PIC, while the receiver of the first PIC may be in communication with the transmitter of a third PIC. The first PIC is not required to receive signals from the second PIC.
In some embodiments, the amplifier and polarization demultiplexing PICs 2340a and 2340b may be configured as depicted in the example of
In some embodiments, the PIC 2510 receives additional sets of wavelengths from off-chip through SM optical fiber 2530. The received light from SM optical fiber 2530 is similarly adjusted for polarization scrambling in the polarization module 2520b. An optical wavelength filter 2514 splits the two sets of wavelengths, providing appropriate routing to the receiver 2516 on the PIC 2510. The wavelength filter 2514 also acts as a circulator and is tunable to send a first set of wavelengths to the transmitter 2512 and a second set of wavelengths different than the first set of wavelengths to the receiver 2516. Alternatively, the positions of the polarization module 2520b and the optical wavelength filter 2514 may be switched because the optical wavelength filter 2514 is polarization independent, with incoming light from the SM optical fiber 2530 first passing through the optical wavelength filter 2514 and then through the polarization module 2520b, as aspects of this technology are not limited in this respect.
In some embodiments, the two paths of light from the PSR 2522 are output to respective tunable delay lines 2524. The respective tunable delay lines 2524 are configured to compensate for polarization mode dispersion (PMD) within the SM optical fibers 2504 and 2530, as PMD is a linear phenomenon, and the TE and TM signals are traveling along the same optical fiber paths. The tunable delay lines 2524 are also configured to retime the two output TE signals so that they may be recombined at combiner 2526 with aligned phases.
In some embodiments, polarization demultiplexing PIC systems 2500 may be combined with an optical flow switch to form a bidirectional communication system. An example of a bidirectional communication system 2600 is depicted in
While the optical flow switch 2610 may be polarization independent, in some embodiments the optical flow switch 2610 may be polarization dependent and include polarization management modules. An example of optical flow switch 2610 including polarization management modules 2613 is shown in
At each input/output of the optical flow switch 2610, wavelength splitters 2612 and 2614 are coupled at either end of the polarization management modules 2613. The wavelength splitters 2612 and 2614 are arranged to select which wavelength of light is passing to the polarization management modules 2613. The wavelength splitters 2612 and 2614 are polarization independent in the configuration shown in the example of
In some embodiments, the polarization management modules 2613 are configured to compensate for polarization scrambling in SM optical fibers, as described in connection with polarization module 2520. Additionally, the polarization management modules 2613 are configured to provide light having the desired polarization to the switching layer 2616 within the optical flow switch 2610.
In some embodiments, the two PICs communicating across the optical flow switch 2610 can include two or more lasers configured to generate the light having the first wavelength 2611a and the light having the second wavelength 2611b. The two PICs can cooperatively determine a configuration where one PIC communicates using the first wavelength 2611a and the other PIC communicates using the second wavelength 2611b.
In some embodiments, three or more PICs may be communicating across the optical flow switch 2610. The system can determine that a first PIC of the three or more PICs will use light having the first wavelength 2611a to send data to a second PIC. Simultaneously, the first PIC may use light having the second wavelength 2611b to transmit data to the third PIC. The second PIC may then transmit data to the first PIC using light having the second wavelength 2611b, and the third PIC may transmit data to the first PIC using light having the first wavelength 2611a. In such embodiments, the optical flow switch 2610 may include polarization management modules 2613 on both optical arms between the wavelength splitters 2612 and 2614, thereby removing the restriction that light of one wavelength travel unidirectionally across the optical flow switch 2610. This arrangement allows for communications between two transceiver PICs that go through multiple flow switches.
The inventors have recognized that the complexity of crossings between waveguides can present challenges in designing the layout of a Benes switch using silicon photonics technology. Accordingly, the inventors have developed techniques for laying out optical flow switches using layer-by-layer or column-by-column cross stamps.
Having thus described several aspects and embodiments of the technology set forth in the disclosure, it is to be appreciated that various alterations, modifications, and improvements will readily occur to those skilled in the art. Such alterations, modifications, and improvements are intended to be within the spirit and scope of the technology described herein. For example, those of ordinary skill in the art will readily envision a variety of other means and/or structures for performing the function and/or obtaining the results and/or one or more of the advantages described herein, and each of such variations and/or modifications is deemed to be within the scope of the embodiments described herein. Those skilled in the art will recognize or be able to ascertain using no more than routine experimentation many equivalents to the specific embodiments described herein. It is, therefore, to be understood that the foregoing embodiments are presented by way of example only and that, within the scope of the appended claims and equivalents thereto, inventive embodiments may be practiced otherwise than as specifically described. In addition, any combination of two or more features, systems, articles, materials, kits, and/or methods described herein, if such features, systems, articles, materials, kits, and/or methods are not mutually inconsistent, is included within the scope of the present disclosure.
Also, as described, some aspects may be embodied as one or more methods. The acts performed as part of the method may be ordered in any suitable way. Accordingly, embodiments may be constructed in which acts are performed in an order different than illustrated, which may include performing some acts simultaneously, even though shown as sequential acts in illustrative embodiments.
All definitions, as defined and used herein, should be understood to control over dictionary definitions, definitions in documents incorporated by reference, and/or ordinary meanings of the defined terms.
The indefinite articles “a” and “an,” as used herein in the specification and in the claims, unless clearly indicated to the contrary, should be understood to mean “at least one.”
The phrase “and/or,” as used herein in the specification and in the claims, should be understood to mean “either or both” of the elements so conjoined, i.e., elements that are conjunctively present in some cases and disjunctively present in other cases. Multiple elements listed with “and/or” should be construed in the same fashion, i.e., “one or more” of the elements so conjoined. Other elements may optionally be present other than the elements specifically identified by the “and/or” clause, whether related or unrelated to those elements specifically identified. Thus, as a non-limiting example, a reference to “A and/or B,” when used in conjunction with open-ended language such as “comprising” can refer, in one embodiment, to A only (optionally including elements other than B); in another embodiment, to B only (optionally including elements other than A); in yet another embodiment, to both A and B (optionally including other elements); etc.
As used herein in the specification and in the claims, the phrase “at least one,” in reference to a list of one or more elements, should be understood to mean at least one element selected from any one or more of the elements in the list of elements, but not necessarily including at least one of each and every element specifically listed within the list of elements and not excluding any combinations of elements in the list of elements. This definition also allows that elements may optionally be present other than the elements specifically identified within the list of elements to which the phrase “at least one” refers, whether related or unrelated to those elements specifically identified. Thus, as a non-limiting example, “at least one of A and B” (or, equivalently, “at least one of A or B,” or, equivalently “at least one of A and/or B”) can refer, in one embodiment, to at least one, optionally including more than one, A, with no B present (and optionally including elements other than B); in another embodiment, to at least one, optionally including more than one, B, with no A present (and optionally including elements other than A); in yet another embodiment, to at least one, optionally including more than one, A, and at least one, optionally including more than one, B (and optionally including other elements); etc.
In the claims, as well as in the specification above, all transitional phrases such as “comprising,” “including,” “carrying,” “having,” “containing,” “involving,” “holding,” “composed of,” and the like are to be understood to be open-ended, i.e., to mean including but not limited to. Only the transitional phrases “consisting of” and “consisting essentially of” shall be closed or semi-closed transitional phrases, respectively.
The terms “approximately,” “substantially,” and “about” may be used to mean within ±20% of a target value in some embodiments, within ±10% of a target value in some embodiments, within ±5% of a target value in some embodiments, within ±2% of a target value in some embodiments. The terms “approximately,” “substantially,” and “about” may include the target value.
This application claims the benefit of U.S. Provisional Application Ser. No. 63/440,332, filed on Jan. 20, 2023, under Attorney Docket No. L0858.70066US01 and entitled “OPTICAL FLOW SWITCHING USING PHOTONIC INTEGRATED CIRCUITS,” and of U.S. Provisional Application Ser. No. 63/503,452, filed on May 19, 2023, under Attorney Docket No. L0858.70075US00 and entitled “MULTI-RETICLE OPTICAL FLOW SWITCH,” and of U.S. Provisional Application Ser. No. 63/511,165, filed on Jun. 29, 2023, under Attorney Docket No. L0858.70075US01 and entitled “OPTICAL FLOW SWITCH WITH CROSSING STAMPS,” each of which is hereby incorporated herein by reference in its entirety.
Number | Date | Country | |
---|---|---|---|
63511165 | Jun 2023 | US | |
63503452 | May 2023 | US | |
63440332 | Jan 2023 | US |